Rumor Detection With Hierarchical Representation on Bipartite Ad Hoc Event Trees

Area of Science:

• Computer Science
• Social Media Analysis
• Artificial Intelligence

Background:

• Social media facilitates rapid information spread, posing challenges for rumor detection.
• Existing methods often overlook propagation topology and author influence, limiting rumor debunking effectiveness.

Purpose of the Study:

• To propose a novel rumor detection model, BAET, that leverages hierarchical representations of bipartite ad hoc event trees.
• To enhance rumor detection by incorporating both propagation structure and author influence.

Main Methods:

• Representing claim posts as bipartite ad hoc event trees (author tree and post tree).
• Utilizing word embedding and feature encoders for respective trees.
• Employing a root-aware attention module for node representation.
• Adopting a tree-like Recurrent Neural Network (RNN) with a tree-aware attention module for structural correlation and tree representation learning.

Main Results:

• BAET demonstrates effectiveness in exploring and exploiting rumor propagation structures.
• The model achieves superior rumor detection performance compared to state-of-the-art methods on public Twitter datasets.

Conclusions:

• The proposed BAET model offers a significant advancement in rumor detection by integrating topological and author-centric features.
• BAET's hierarchical representation on bipartite ad hoc event trees proves effective for accurate rumor identification and debunking.